VOLUME 66, NUMBER 8 PHYSICAL REVIEW LETTERS 25 FEBRUARY 1991

Formation of Naked Singularities: The Violation of Cosmic Censorship

Stuart L. Shapiro and Saul A. Teukolsky
Center for Radiophysics and Space Research and Departments of Astronomy and Physics,
Cornell University, Ithaca, New York 14853
(Received 7 September 1990)

We use a new numerical code to evolve collisionless gas spheroids in full general relativity. In all cases
the spheroids collapse to singularities. When the spheroids are su%ciently compact, the singularities are
hidden inside black holes. However, when the spheroids are su%ciently large, there are no apparent hor-
izons. These results lend support to the hoop conjecture and appear to demonstrate that naked singulari-
ties can form in asymptotically flat spacetimes.

PACS numbers: 04.20.Jb, 95.30.Sf, 97.60.Lf

It is well known that general relativity admits solu-
tions with singularities, and that such solutions can be
produced by the gravitational collapse of nonsingular,
asymptotically flat initial data. The cosmic censorship
hypothesis' states that such singularities will always be
clothed by event horizons and hence can never be visible
from the outside (no naked singularities). If cosmic cen-
sorship holds, then there is no problem with predicting
the future evolution outside the event horizon. If it does
not hold, then the formation of a naked singularity dur-
ing collapse would be a disaster for general relativity
theory. In this situation, one cannot say anything precise
about the future evolution of any region of space con-
taining the singularity since new information could
emerge from it in a completely arbitrary way.
Are there guarantees that an event horizon will always
hide a naked singularity? No definitive theorems exist.
Counterexamples are all restricted to spherical symme-
try and typically involve shell crossing, shell focusing, or
self-similarity. Are these singularities an accident of
spherical symmetry?
For nonspherical collapse Thorne has proposed the
hoop conjecture: Black holes with horizons form when
and only when a mass M gets compacted into a region
whose circumference in every direction is C 4aM. If
the hoop conjecture is correct, aspherical collapse with
one or two dimensions appreciably larger than the others
might then lead to naked singularities.
For example, consider the Lin-Mestel-Shu instability
for the collapse of a nonrotating, homogeneous spheroid
of collisionless matter in Newtonian gravity. Such a
configuration remains homogeneous and spheroidal dur-
ing collapse. If the spheroid is slightly oblate, the con-
"
figuration collapses to a "pancake, while if the spheroid
is slightly prolate, it collapses to a spindle. While in both
cases the density becomes infinite, the formation of a
spindle during prolate collapse is particularly worrisome.
The gravitational potential, gravitational force, tidal
force, and kinetic and potential energies all blow up.
This behavior is far more serious than Inere shell cross-
ing, where the density alone becomes momentarily infi-
nite. For collisionless matter, prolate evolution is forced
to terminate at the singular spindle state. For oblate
evolution, the matter simply passes through the pancake
state, but then becomes prolate and also evolves to a
spindle singularity.
Does this Newtonian example have any relevance to
general relativity? We already know that inftnite cylin-
ders do collapse to singularities in general relativity, and,
in accord with the hoop conjecture, are not hidden by
event horizons. But what about finite configurations in
asymptotically flat spacetimes?
Previously, we constructed an analytic sequence of
momentarily static, prolate, and oblate collisionless
spheroids in full general relativity. We found that in the
limit of large eccentricity the solutions all become singu-
lar. In agreement with the hoop conjecture, extended
spheroids have no apparent horizons. Can these singu-
larities arise from the collapse of nonsingular initial
data? To answer this, we have performed fully relativis-
tic dynamical calculations of the collapse of these
spheroids, starting from nonsingular initial configur-
ation.
We find that the collapse of a prolate spheroid with
sufticiently large semimajor axis leads to a spindle singu-
larity without an apparent horizon. Our numerical com-
putations suggest that the hoop conjecture is valid, but
that cosmic censorship does not hold because a naked
singularity may form in nonspherical relativistic collapse.
—
Wurnerical code. Our numerical code solves Ein-
stein's equations for the evolution of nonrotating, col-
lisionless matter in axisymmetric spacetimes. The field
equations are expressed in 3+1 form following Arnowitt,
Deser, and Misner. We use maximal time slicing and
isotropic spatial coordinates. The field equations reduce
to a coupled set of nonlinear hyperbolic and elliptic par-
tial diff'erential equations which we solve by finite dif-
ferencing. The matter equations are the geodesic equa-
tions in the self-consistent gravitational field. The nu-
merical treatment is a mean-field particle simulation
scheme that solves the Vlasov equation in general rela-
tivity. It is an extension of our previous relativistic treat-

994 1991 The American Physical Society

VOLUME 66, NUMBER 8 PHYSICAL REVIEW LETTERS 25 FEBRUARY 1991

ment for spherical spacetimes and our Newtonian
method ' for axisymmetric configurations. The code can
handle matter velocities approaching the speed of light
and strong gravitational fields, including black holes. It
is designed to treat cases in which the collisionless matter
collapses to a singularity: Specifically, oblate collapse to
flat pancakes and prolate collapse to thin spindles.
We have carried out a large battery of test-bed calcu-
lations to ensure the reliability of the code. These tests
included the propagation of Iinearized analytic quadru-
pole waves with and without matter sources and non-
linear Brill waves in vacuum spacetimes; maintaining
equilibria and identifying the onset of instability for
2— | i I
(
1 I t I
i
I f
spherical equilibrium clusters; reproducing Oppenheim-
er-Snyder collapse of homogeneous dust spheres and
Newtonian collapse of homogeneous spheroids. We '
constructed a number of geometric probes to diagnose
the evolving spacetime. We tracked the Brill mass and
outgoing radiation energy flux to monitor mass-energy
conservation. To confirm the formation of a black hole,
we probed the spacetime for the appearance of an ap-
parent horizon and computed its area and shape when it
was present. To measure the growth of a singularity, we
computed the Riemann invariant I— = R,p»R ~~ at every
spatial grid point. To test the hoop conjecture, we com-
puted the minimum equatorial and polar circumferences
outside the matter.
Typical simulations were performed with a spatial grid
I I
i
I I I I

of 100 radial and 32 angular zones, and with 6000 test

t/M =0
particles. A key feature enabling us to snuggle close to
singularities was that the angular grid could fan and the
i. 5— radial grid could contract to follow the matter.
Collapse of collisionless spheroids. We followed the —

to— t/M =0
0.5 —-
8—
0 I
j

2—
I I
[

6—
t/M = 7, 7—
x
a 4—
t, 5—

0-'i i. i. l. ~ i t I

10— t/M = 23
0

05 1 1.5 2
6—
Equator V)
X
FIG. 1. Snapshots of the particle positions at initial and late
times for prolate collapse. The positions (in units of M) are 4
projected onto a meridional plane. Initially the semimajor axis
of the spheroid is 2M and the eccentricity is 0.9. The collapse
proceeds nonhomologously and terminates with the formation
of a spindle singularity on the axis. However, an apparent hor- 0 I
'.
I I I i I I I I 1

izon (dashed line) forms to cover the singularity. At t/M =7.7 0 2 4 6 8 10

its area is A/16@M =0.98, close to the asymptotic theoretical
limit of 1. Its polar and equatorial circumferences at that time
are C ~~//4~M = 1.03 and C,q /4rrM =0.91. At later times
these circumferences become equal and approach the expected
theoretical value 1. The minimum exterior polar circumfer-
ence is shown by a dotted line when it does not coincide with
the matter surface. Likewise, the minimum equatorial cir-
cumference, which is a circle, is indicated by a solid dot. Here
Cg'"/4aM=0. 59 and CPg/4zrM=0. 99. The formation of a
black hole is thus consistent with the hoop conjecture.
Equator
FIG. 2. Snapshots of the particle positions at initial and
final times for prolate collapse with the same initial eccentrici-
ty as Fig. 1 but with initial semimajor axis equal to 10M. The
collapse proceeds as in Fig. 1, and terminates with the forma-
tion of a spindle singularity on the axis at t/M=23. The
minimum polar circumference is C'Pl", /4+M=2. 8. There is no
apparent horizon, in agreement with the hoop conjecture. This
is a good candidate for a naked singularity, which would vio-
late the cosmic censorship hypothesis.

995
VOLUME 66, NUMBER 8 PHYSICAL REVIEW LETTERS 25 FEBRUARY 1991

collapse of nonrotating prolate and oblate spheroids of
various initial sizes and eccentricities. The matter parti-
cles are instantaneously at rest at t=0 and the con-
figurations give exact solutions of the relativistic initial-
value equations. In the Newtonian limit, these spher-
oids reduce to homogeneous spheroids. When they are
large (size»M in all directions), we confirm that their
evolution is Newtonian. ' to half its peak value at r =4. 5M inside the matter and
Figure 1 shows the fate of a typical prolate config-
uration that collapses from a highly compact and relativ-
istic initial state to a black hole. Note that in isotropic
coordinates a Schwarzschild black hole on the initial
time slice would have radius r =0.5M, corresponding to
a Schwarzschild radius rg =2M. Figure 2 depicts the
outcome of prolate collapse with the same initial eccen-
tricity but from a larger semimajor axis. Here the con-
figuration collapses to a spindle singularity at the pole
without the appearance of an apparent horizon. The
spindle consists of a concentration of matter near the
axis at r =5M. Figure 3 shows the growth of the
Riemann invariant I at r =6. 1M on the axis, just outside
the matter. Before the formation of the singularity, the
typical size of I at any exterior radius r on the axis is
-M /r «1.
With the formation of the spindle singu-
I
larity, the value of rises without bound in the region
near the pole. The maximum value of determined by
our code is limited only by the resolution of the angular
grid: The better we resolve the spindle, the larger the
I
value of we can attain before the singularity causes the
code (and spacetime) to break down. Unlike shell-
crossing singularities, where I blows up in the matter in-
terior whenever the matter density is momentarily
infinite, the singularity also extends outside the matter
lO I I I
I
I i t I
I
I c
beyond the pole at r =5.8M (Fig. 4). In fact, the peak
I
value of occurs in the vacuum at r =6. 1M. Here the
exterior tidal gravitational field is blowing up, which is
not the case for shell crossing. The absence of an ap-
parent horizon suggests that the spindle is a naked
singularity.
When our simulation terminates, I along the axis falls
r=6. 7M outside the matter. The singularity is not a
point. Rather it is an extended region which includes the
matter spindle, but grows most rapidly in the vacuum ex-
terior above the pole. A t =const slice has a spatial
metric
2
+ 2dr2+Ppde++2r2 sin 2gdp2
In Aat space A =8=1. At the termination of the simu-
" lation these quantities have a modest maximum value
2 =8= 1.7, which occurs at the origin. They decrease
monotonically outwards, reaching unity at large dis-
tances. However, it is their second derivatives that con-
I
tribute to and these blow up. While A and 8 steadily
grow with time, I diverges much more rapidly. The be-
havior is similar to the logarithmic divergence of the
metric in the analytic prolate sequence of Ref. 6. We
emphasize that the above characterization of the singu-
I larity and the behavior of the metric is dependent on the
time slicing and may be different for other choices of
time coordinate. In principle, the spindle singularity
might first occur at the center rather than the pole with a
different time slicing.
The absence of an apparent horizon does not neces-
sarily imply the absence of a global event horizon, al-
though the converse is true. Because such singularities
cause our numerical integrations to terminate, we cannot
map out a spacetime arbitrarily far into the future,
I I
i
I I I I
I
I t I I
I
I I I
which would be necessary to completely rule out the for-
t64 mation of an event horizon. However, we do not think
this is at all likely: For collapse from an initially com-
pact state (Fig. 1), outward null geodesics turn around

|t 32

)0
I I I I

5
I I I I I

l0
I

)5
I

20 25 0
t/M Q~
FIG. 3. Growth of the Riemann I (in units of invariant
M ) vs time for the collapse shown Fig. 2. The simulation in
was repeated with various angular grid resolutions. Each curve I
FIG. 4. Profile of in a meridional plane for the collapse
is labeled by the number of angular zones used. We use dots to shown in Fig. 2. For the case of 32 angular zones shown here,
show where the singularity has caused the code to become I
the peak value of is 24/M and occurs on the axis just outside
inaccurate. the matter.

996
VOLUME 66, NUMBER 8 PHYSICAL REVIEW LETTERS
near the singularity. For collapse from large radius, by
contrast (Fig. 2), outward null geodesics are still propa-
gating freely away from the vicinity of the singularity up
to the time our integrations terminate. It is an interest-
ing question for future research whether any time slicing
can be found which will be more effective in snuggling
'
up to the singularity without actually hitting it. Such
a slicing would enable one to confirm that all outward
null geodesics propagate to large distances.
Further evidence for the nakedness of the singularity
is the similarity of the spindle singularity to the infinite
cylinder naked singularity. ' In both cases the proper
length of a given segment of matter along the axis grows
slowly, while its proper circumference and surface area
shrink to zero much more rapidly. Also, the singularity
is an extended region along the axis and not just a point.
We have also followed the collapse of an initially ob-
late configuration with the same initial eccentricity and
semimajor axis as Fig. 2. Following pancaking, it over-
shoots, becomes prolate, and forms a black hole. At the
time our integrations terminate, we find that 'Eppes 8 q'"
=0.85(4~M ).
All of the above results are consistent with the hoop
conjecture. When black holes form, the minimum polar
and equatorial circumferences satisfy C™n~4aM. Con-
versely, when naked singularities form, the minimum po-
lar circumference is much bigger than this value. In all
cases where an apparent horizon forms, its area satisfies
to within numerical accuracy A ~ 16aM, as required
theoretically. ' In every case we find that gravitational
radiation carries away a negligible fraction («1%) of
the total mass energy by the time a black hole or naked
singularity forms.
In conclusion, we have presented numerical evidence
that the hoop conjecture is a valid criterion for the for-
mation of black holes during nonspherical gravitational
collapse. We have also found numerical candidates for
the formation of naked singularities from nonsingular in-
itial configurations. These examples are in contrast with
any cases of singularities which may arise during spheri-
cal collapse. There the exterior spacetime is always the
Schwarzschild metric and the Riemann invariant is al-
ways exactly 48M /rP, which is finite outside the matter.
In spherical collapse the singularities can thus only occur
inside the matter. Here the singularities extend above
the pole into the vacuum exterior. These examples sug-
gest that the unqualified cosmic censorship hypothesis
cannot be valid.
While the matter treated here has kinetic pressure, it
is collisionless, not fluid. We do not regard the collision-
al properties of the matter as crucial: First, the forma-
tion of naked singularities should not depend on the par-
ticular details of the fundamental interactions affecting
matter at high densities. The gravitational field equa-
tions alone should be sufficient to rule out naked singu-
larities, at least in the vacuum exterior, for true cosmic
censorship. Second, collisional eAects may even ac-
25 FEBRUARY 1991
celerate the formation of singularities via relativistic
"pressure regeneration. ' "
The examples considered here have no angular
momentum. The presence of angular momentum could
prevent an infinitesimally thin spindle singularity from
forming on the axis. Yet we know that a small amount
of angular momentum does not prevent the formation of
a singularity when a Kerr black hole forms. In any case,
it would be disturbing if a collapsing configuration could
become arbitrarily close to singular for arbitrarily small
angular momentum.
This research was supported in part by National Sci-
ence Foundation Grants No. AST 87-14475 and No.
PHY 90-07834 at Cornell University. Computations
were performed on the Cornell National Supercomput-
er Facility. We thank A. Abrahams, M. Choptuik,
C. Evans, and L. S. Finn for helpful discussions.
'R. Penrose, Riv. Nuovo Cimento 1 (Numero Special), 252
(1969).
See, e.g. , D. S. Goldwirth, A. Ori,and T. Piran, in Frontiers
in Numerical Relativity, edited by C. R. Evans, L. S. Finn,
and D. W. Hobill (Cambridge Univ. Press, Cambridge, 1989),
p. 414, for discussion and references.
K. S. Thorne, in Magic Without Magic: John Archibald
Wheeler, edited by J. Klauder (Freeman, San Francisco,
1972), p. 1.
4C. C. Lin, L. Mestel, and F. H. Shu, Astrophys. J. 142,
1431 (1965).
5C. W. Misner, K. S. Thorne, and J. A. Wheeler, Gravita-
tion (Freeman, San Francisco, 1973), p. 867.
T. Nakamura, S. L. Shapiro, and S. A. Teukolsky, Phys.
Rev. D 38, 2972 (1988).
7R. Arnowitt, S. Deser, and C. W. Misner, in Gravitation:
An Introduction to Current Research, edited by L. Witten
(Wiley, New York, 1962), p. 227.
C. R. Evans, Ph. D. thesis, The University of Texas at Aus-
tin, 1984 (unpublished).
S. L. Shapiro and S. A. Teukolsky, Astrophys. J. 298, 34
(1985); 298, 58 (1985); 307, 575 (1986); Astrophys. J. (Lett. ),
292, L41 (1985).
'OS. L. Shapiro and S. A. Teukolsky, Astrophys. J. 318, 542
(1987).
''We searched for both a single global horizon centered on
the origin as well as a small disjoint horizon around the singu-
larity in each hemisphere.
' Maximal slicing apparently does not hold back the forma-
tion of prolate spindle singularities. For prolate spheroids the
Newtonian potential diverges only logarithmically as the ec-
centricity 1, which may explain why a does not plummet
precipitously near a spindle.
'3See, e.g. , D. M. Eardley, Phys. Rev. D 12, 3072 (1975).
' See, e.g. Ref. 5, p. 605. Moreover, there is at least one ten-
,
tative numerical example [T. Nakamura and H. Sato, Prog.
Theor. Phys. 68, 1396 (1982); T. Nakamura, K. Oohara, and
Y. Kojima, Prog. Theor. Phys. Suppl. 90, 57 (1987)] of prolate
Quid collapse (adiabatic index I 2) that appears to be evolv-
ing to a singular state without the formation of an apparent
horizon.
997